I am currently working on a journal regarding the the ever-elusive topic of human attractiveness. I have gathered a lot of photographic data, but I'm currently facing a challenge. I can't seem to find an effective digital image processing software solution for evaluation of facial attractiveness. I have looked into different morphometrics software solutions but I can't find the right one. I have also considered adapting facial recognition software to my needs but it seems they are designed to measure facial landmarks over large distances and with not very high accuracy. I have a lot of high definition photographs and I am searching for a very accurate method to analyze them. Maybe there is a solution I can adapt to my needs. Any suggestions would be greatly appreciated!
Hi,
I might help you with this problem. Do you have any labels for attractiveness in your dataset?
Hi Alexander,
A very interesting problem. We just added a facial feature finding tool to our new software MIPAR (http://mipar.us/final-release). Combined with all the other image processing and quantification tools it has, it may just have the power and flexibility you need to help your study. Happy to work with you on it.
Cheers,
John
The very basic idea for computing facial attractiveness is to analyze measure among facial features and quantify the facial symmetry. Those measures may be easily performed using fiducial points.
We have lately introduced a neural-network based classifier that is able to extract 40 facial attributes from an input image. One of these attributes is "Attractive", see: https://arxiv.org/abs/1603.07027 However, the classification accuracy of the "Attractive" attribute is amongst one of the worst. Hence, I am not sure how well attractiveness can be classified in the first place.
I have read some older articles about classifying attractiveness (or beauty) -- most of them only talking about female facial beauty, for example: 252513124_Automated_classification_of_female_facial_beauty_by_image_analysis_and_supervised_learning or http://ieeexplore.ieee.org/document/5346057
I am not sure if any of these would help you.
I always feel that holistic approaches (some CNNs for example) are way better than parametric approaches like detecting some fiducial points and calculating some distances accordingly.
The labels are crucial because the system work in harmony with some observational data.
You might consider examining distance from the facial average (the closer to the average, the more attractive). See Langlois & Roggman (1990) and Potter, Corneille, Ruys, & Rhodes (2007).
Jennifer, I intuitively would agree with you. But then a 'stunning beauty' would not exist; only 'mediocre (average) beauties'. Kind of sad...
- Badges
- Science topic